1 CHAPTER MEANING AND REFERENCE: SOME CHOMSKIAN THEMES ROBERT J. STAINTON THIS chapter introduces three arguments that share a single conclusion: that a comprehensive science of language cannot (and should not try to) describe relations of semantic reference, i.e. word-world relations. Spelling this out, if there is to be a genuine science of linguistic meaning (yielding theoretical insight into underlying realities, aiming for integration with other natural sciences), then a theory of meaning cannot involve assigning external, real-world, objects to names, nor sets of external objects to predicates, nor truth values (or world-bound thoughts) to sentences. Most of the chapter tries to explain and defend this broad conclusion. The chapter also presents, in a very limited way, a positive alternative to external-referent semantics for expressions. This alternative has two parts: first, that the meanings of words and sentences are mental instructions, not external things; second, as Strawson (1950) stressed, that it is people who refer (and who express thoughts) by using words and sentences, and wordlsentence meanings play but a partial role in allowing speakers to talk about the world. I am very grateful to Ash Asudeh, Alex Barber, Akeel Bilgrami, Andrew BottereU, Andy Brook, Ray Elugardo, Corinne Iten, Ernie Lepore, David Matheson, Julius Moravcsik, Paul Pietroski, Aryn Pyke, Jim McGilvray, Ray Jackendoff, and Catherine Wearing for comments on earlier drafts. Work on this chapter was supported financially by grants from the Canada Research Chairs program, the Ontario Ministry of Science, Energy and Technology, and the Social Science and Humanities Research Council of Canada.
3 ' features of a naturally occurring phenomenon, always looking forward to eventual integration with the core natural sciences.' Second, just as the sciences in general do not feel overly constrained by what "folk theorizing" suggests, or by how ordinary folks use words, for the Chomskian, linguistics and psychology are allowed to (indeed, they are expected to) put common sense conceptions aside, and to use terms in specialized ways, etc. Indeed, as Chomsky (1993: 25) rightly suggests, modern science gets going precisely when one is willing to be surprised by what are, from the perspective of common sense, "simple phenomena": e.g. that rocks fall, that people get sick and die, that a phrase is ambiguous, et~.~ Third, since the sciences in general take their evidence wherever they can find it, there can be, for the methodological naturalist, no a priori restrictions on evidence in psychology or linguistics. On these grounds alone, much that has become conventional wisdom in the study of language-whether deriving from common sense talk, or from abstract philosophizing-has to be re-evaluated carefully. If one studies the mind and language this way, taking preconceptions with a grain of salt, scientific inquiry into the salient natural object reveals-continues this line of thought-two less immediate implications. First, that the mature speakerlhearer's mind contains far more information than can be gleaned from the environment. This is the finding of the poverty of the stimulus. The most natural explanation of this finding, and the one that any unbiased scientist would immediately pursue, is that the human mind, including in particular the part of it responsible for language, benefits from a substantial innate endowment. A different though related hypothesis that emerges in this scientific endeavor is that the mind is divided, by nature, into a series of specialized faculties-rather than being, say, a homogenous "cognitionllearning machine". This is the empirical hypothesis of modularity, with the language faculty being a case in point.4 For the methodological naturalist, that some people find these latter results initially counterintuitive carries no real weight: after all, one should no more trust "intuitions'' about brain structure and brain development than one should trust intuitions about the development and structure of the liver. See Chomsky, 1992a: 19, Chomsky, 1992b: 53 and Chomsky, For the idea of "knowledge that would suffice for interpretation", see Davidson, 1976, which builds on Foster For a very balanced comparison of this and other philosophical projects with Chomsky's naturalist one, see B. Smith, A trenchant critique of the former projects may be found in Antony This does not, of course, entail that Moore-style "common sense propositions"-e.g. "that there exist now both a sheet of paper and a human hand (Moore, 1939: 165)-should be rejected as false. As will emerge below, the methodological naturalist perspective does not conflict with common sense views about particular matters; rather, it pursues a different path entirely. As Chomsky (1995b: 138-9) writes: "It is not that ordinary discourse fails to talk about the world, or that the particulars it describes do not exist, or that the accounts are too imprecise. Rather, the categories used and principles invoked need not have even loose counterparts in naturalistic inquiry". It's worth noting that Chomsky employs a different notion of 'module' than, say, Fodor, (1983) does. Also, some read Chomsky as merely stipulating that linguistics, in his sense, studies what he labels "I-language": the intensionally characterized rules internal to the individual language faculty. This understates his claims. Chornsky's point, I take it, is that an unbiased methodological naturalist will study I-language, rather than other possible constructs, because the I-language construct turns out to correspond to a real aspect of the natural world that emerges in careful inquiry, whereas other constructs do not. Again, see Bezuidenhout, this volume, for more.
5 y-- MEANING AND REFERENCE: SOME CHOMSKIAN THEMES 917 reasons. First, not all true sentences mean the same thing, nor do all false sentences mean the same thing: 'Five is larger than two' is not synonymous with 'France is in ~urope', though both are true. So truth values are not finely grained enough to be meanings. Second, it's not obvious that sentences are even the right kind of thing to be truelfalse. Certainly many sentences are explicitly context sensitive: 'He bought that yesterday', for instance, is at best only true relative to an assignment of values for 'he', 'that', and 'yesterday'. Thus, this sentence just isn't the sort of entity to even have a truth value tout court. To overcome the first problem, one can take sentences to denote not truth values but propositions made up of objects and ~roperties; or one can take them to have a truth value as referent, but a Fregean Thought as sense; or one can take the meaning of a sentence to be truth-conditions. such are the treatments of natural language sentences proposed by the contemporary ~ussellian, Fregean, and Tarskian respectively. These distinctions about sentence meanings, though important for other purposes, can be glossed over here: in the present chapter, I will speak vaguely of indicative sentences "expressing thoughts". To overcome the second problem, that natural language sentences contain contextsensitive items like 'he', 'that 1, and 'yesterday', one can say that such sentences express not thoughts exactly, but "proto-thoughts": something which is true or false relative to a set of contextual parameters (time, place, speaker, addressee). Such, then, is the first plank of the view to be reje~ted.~ Typically added to this idea, in "mainstream" philosophy of language, is the assumption that languages are the common property of a whole community, such that the symbolic items (words, predicates, sentences) are all public property. Languages, on this view, exist independently of speakers; and, being a public entity, each speaker typically has only a partial grasp of hislher shared language. This is the second plank of the view to be rejected. (For more on the Chomskyan target, see Bezuidenhout, this volume.) THE THREE NEGATIVE ARGUMENTS The Radical "Argument from Ontology" Having clarified what the target is, and what makes the critiques in question broadly "Chomskian", I turn to the negative arguments. The view at issue invokes relations between public linguistic items on the one hand (names, predicates, sentences) and worldly items on the other (external objects, sets, world-bound thoughts). Obviously, then, the relata must be able to stand in the Famously, the philosopher-logicians who are taken to be the grandfathers of this tradition-i.e. Frege, Russell, and Tarski-explicitly disavowed the idea that natural languages, in all their messy detail, could be treated this way. It was their contemporary philosophical followers, most notably Davidson (1967) andmontague (1974), who took the analogy between logical and natural languages literally. Chomsky et nl. emphatically agree with the grandfathers, and disagree with their contemporary heirs.
6 requisite relations, including in particular (something like) the denotation relation. But, patently, the relata can't do this job if they aren't real. The first negative argument questions whether they are. The widespread idea, to be questioned here, is that there are public signs, of shared languages, available to have referents: words, phrases and sentences that belong to languages like English, Urdu, and Swahili. In what follows, I will introduce three worries about this idea, to give the flavor of the thing. (One could easily raise many more.) The first problem has to do with individuation of words given variation. Crucially, as actual working linguists are wont to note, the way we divide up languages in common sense, and in much philosophical theorizing, does not actually correspond to any robust divide. One speaks of "Chinese" as a single language, despite the fact that its two largest "dialects", Mandarin and Cantonese, are not mutually intelligible. In contrast, we call Spanish, Portuguese, and Italian different languages, rather than speaking of several dialects of Romance, just as we treat Swedish and Danish as different languages-this despite the fad that they are much more similar to each other, and far closer to mutual intelligibility, than the "dialects of Chinese" are. The only semi-robust divide here is mutual intelligibility and, as noted, "languages" are not divlded along those lines. One might reasonably reply that this worry can be overcome by thinking of words as belonging to dialects, not languages. But that won't really help, since what counts as a dialect is equally peculiar: Canadian English is supposed to be a single dialect, despite the many differences between speakers in urban centers and rural areas, and differences among the East, Central Canada, and the West; it also is supposed to be a different dialect than what is spoken in, say, Ohio. Clearly, we slice things as we do-both "languages" and "dialects n -not because of any robust linguistic divide, but because of colonial history, similar writing systems, shared canonical works of literature, present military might, arbitrary national boundaries, religious differences. and so on. That, and not "nature's ioints", is what makes it the case that people "speak the same languageldialect". As Chomsky puts it, "This idea [of a common public language] is completely foreign to the empirical study of language... What are called "languages" or "dialects" in ordinary usage are complex amalgams determined by colors on maps, oceans, political institutions and so on, with obscure normative-teleological aspects" (1993: 18-19). (See also Chomsky, 1992b: 48; Chomsky, 1995b: 155 ff, and Bezuidenhout, this volume.) 1 1 Instead of public languagesldialects, the real objects that one hds are (i) individual idiolects, (ii) sets of idiolects that share some non-obvious underlying parametric feature (e.g. having complements falling after heads), and (iii) the universally shared language faculty. None of these, however, corresponds even remotely to "public languages" like English and Urdu. Now, to come to the problem of immediate interest here, if the boundaries around "languages" (or"dialects") don't reflect an objective difference in kind, what individuates a word in a language? What makes it the case, for instance, that distind pronunciations are pronunciations "of the same word", if there aren't really object- 3 ively distinct languages? To take an example, why are 'fotografer' (said in Bombay)
7 / E MEANING AND REFERENCE: SOME CHOMSKIAN THEMES 919 k 6 and 'fo~~hgrafer' (said in Toronto) the same word, yet 'fotografo' (said in Buenos Aires) is not the same word as the former two? We are wont to say that there are wo words here-the "English" word and the "Spanish" word-not three words. BU~ this won't do, if "English" isn't objectively real: after all, all three differ in pronunciation. For that matter, even within a single country, or a single part of a country, there can be many "different pronunciations of the same word. So, as noted, appeal to local dialects isn't likely to help either. For instance, even within the Eastern united States, there are many pronunciations of 'Harvard'. More than that, children don't pronounce things the same way adults do, women don't pronounce things the same as men, and so on. Given variation, there thus seems to be no good reason to count public words the way common sense wishes to: we can't put aside the differences on reasonable grounds. A natural reply to this first problem about counting words is that a dialect, or a language, is the symbol system shared by community such-and-such. But this reply is quite unhelpful, for at least two reasons. First, a specific worry: it's not possible to individuate the right community except by appeal to shared language. In particular, as we saw above, mutual intelligibility won't allow us to distinguish groups along lines that correspond to "languages". What "the community" for whom 'foto- GRAFer' and 'fotahgrafer' are supposedly one word really have in common, and what distinguishes this "community" from others, is that everyone in it speaks English! A broader worry is that communities are no more "robust" than languages turned out to be. So even if one could divide languages in terms of which communities used them, this still wouldn't yield the kind of robust divide that the methodological naturalist demands. There is a second reason why it is hard to individuate "public language words", beyond the problem of individuation in the face of across-speaker variation. It has to do with how to count words even granting the existence of languagesldialects. To pick an example essentially at random, is there one word 'forge' which has multiple meanings: create a fraudulent imitation, shape by heating in a fire and hammering, and firnace or hearth for melting or refining metal? Or are there three words, one for each meaning?' And, even restricting ourselves to one of the meanings, are 'forged', 'forges', and 'forging' wholly different words, or are they merely variations on the same word? What about the tensed verb 'forged', as in 'He forged the document', the past participle, as in 'He has forged many documents' and the adjective as in 'A forged document'? Are they precisely the same word, wholly different words, or variations on a single word? Also, if there is just one word here, or variations on it, what is that word? Rather than calling out for discovery of something real, these seem matters of decision. In light of these questions about individuation, both across and within a "dialect", one can readily doubt that there is any such thing as "words in English", "sentences ' And note the potential problem of circularity, if one does individuate words by their meanings. Meaning, recall, is supposed to derive from having two things stand in a relation. But now it turns out that one of the relata, on the "word" side, is individuated in terms of the other.
8 in Swahili", and so forth. To echo Quine, one might insist that there can be no entity without identity. But if there are no such objects, there patently cannot be a science of word-world relations that pairs "public words and sentences" with worldly objects, sets, and proto-tho~ghts.~ (Granted, for all that's been said so far, there might be other things that can be paired with external objects: morphemes of an individuars mental lexicon, for instance. But this possibility offers little solace to the kind of theorist that Chomskians are targeting.) I said that I would introduce three problems about words. The third one involves issues about language norms. Though almost universally used among "English speaking" children, 'broke&, 'runned', 'swirnrned' and so on are not "words in English"? Or again, despite its constant appearance in speech and writing, there isn't supposed to be a word in English that means it is to be hoped that, and is pronounced 'hopefully'. On the other hand, supposedly there is an English word pronounced 'ke-naw', because that's how Shakespeareans said 'know'; and there is, according to my Oxford dictionary, an English word 'peavey', even though almost no one would recognize it as such. These latter items aren't used, but they are "English words"; the fonner items are used, but aren't "English words". Clearly, what rules these words in or out is not how people do speak, but rather something about how they should speak. It's at least not obvious how there can really be such things, to stand in objective relations with external objects, sets thereof, and so on. I pause to quickly summarize, before introducing a major objection to this line of argument. Because there is no objective way to individuatelcount words (across or within a "dialect"), and because what makes something a shared, public word, if there really were any, would need to appeal to "ought" rather than "is", the Chomskian concludes that there aren't really any "public words". But then there cannot be a comprehensive science of language that pairs words (and sentences) with external things. Such is the radical argument from ontology. A natural reaction to the claim that words (e.g. 'forged', 'photographer' and 'Harvard') are not real objects is perplexed disbelief that the claim has been seriously made. Surely it's just obvious that words exist. Besides, if an argument is needed, there is this: here we are discussing the various pronunciations of the words 'Harvard' and 'photographer'; and above it was said that the word 'peavey' exists because of norms. But how can something which doesn't exist have different For those familiar with Chomsky's (1986) terminology, the central point may be put like this: public language words/sentences are part of the E-language picture, and the methodological naturalist must eschew E-language as not a suitable candidate for scientific study. See Bezuidenhout, this volume, for discussion. This point relates to another one that Chomsky regularly raises. Public language approaches are at a loss to find a "thing" which children under, say, six years of age know. They don't yet "know English" (or Swahili, or Urdu, or...). Indeed, there is no "public language" which they know at this age. But then how, positively, are we to describe the state of their minds? It seems absurd that we can only make the negative claim: i.e. that they do not yet know English (or Swahili, or Urdu, or...), but are on the way to doing so. Note too, how well such children communicate. This puts the lie to the idea that having "a shared public language" is genuinely necessary for communication.
9 MEANING AND REFERENCE: SOME CHOMSKIAN THEMES 921 pronunciations--as the argument itself grants that 'Hanard' and 'photographer' do? And surely, if something exists because of norms, then it exists. (Indeed, we seem to infer the non-existence of words on the grounds that 'peavey' exists! That's patently absurd.) Our discussion thus seems to give rise to paradox. Given the obviousness of the existence of words, and the paradoxes that quickly arise from denying their existence, it's hard to see how it could be suggested, at least with a straight face, that public words do not really exist. There are several replies to this natural worry. On the one hand, one can agree that these things are real enough, but go on to question whether there could be a science that treated of them. Where by 'a science' is meant, to repeat, rather more than "any inquiry that is both theoretical and empirical". As hinted at the outset, 'science' in the context of methodological naturalism means, at a minimum, seeking explanatory insight; which in turn entails positing underlying realities, and aiming for integration with the core natural sciences. Many things exist which are not subject to scientific investigation, in this sense. This concessive reply will be considered at length in the second negative argument. To anticipate briefly here, the core idea is that the standard for being a "real object" has been set too high in the discussion above. It's not just public languages and words, but corporations, songs, countries, universities, national dishes, hair styles, TV shows, etc., that won't really exist given this over-high standard. Indeed, it's arguable on similar grounds that none of us exist: to see why, think of the enormous puzzles about how to individuate persons. A natural alternative view, which doesn't set the standard so high, is that perfectly real objects can be quite hard to individuatelcount, and can be norm-bound. They need not require a "robust divide", but can rather be objectively different only in degree, with human interests setting the kind-divide between them. One could thus allow that there is such a thing as English (and other public tongues), and that the nature of English and the wordslsentences in it depend on a host of complex relations (political, military, historical, religious, etc.)-including even explicitly normative ones having to do with "correct speech". Adding, goes the reply, that this does not make English and its elements unreal. Personally, I think there is something very importantly right about this. Still, the key point that will re-emerge in negative argument two is that, even granting this, one is hard-pressed to rescue the idea that a genuine science of language can, or should try to, describe word-world relations. Indeed, the account proposed of what makes words and languages real-e.g. that their individuation rests on norms, quirky anthropocentric interests, and a complex mess of other things-pretty much ensures that they will not be scientifically tractable. More than that, if that's what makes something a "word", it's not even plausible that "public word" will be an idealization that will be of any use in science. As Chomsky puts the general point: Such informal notions as Swedish-vs.-Danish, norms and conventions, or misuse of language are generally unproblematic under conditions of normal usage, as is "near New York" or "looks like Mary". But they can hardly be expected to enter into attempts to reach theoretical understanding. (1993: 20)
10 922 ROBERT I. STAINTON - As I say, this concessive reply will be elaborated at length in the next section. But there are non-concessive replies too, which try to defend the radical version of the "argument from ontology" according to which one side of the supposed relations (i.e. the public words/sentences) just do not exist at all. Let me introduce a couple of those replies here. That there are no public words or languages strikes us as absurd, but-goes the first reply-that is because we are taken in by an illusion of some sort. Part of the concept of "public word", the argument would go, is that the things in question are "out there", the shared property of many. They are not inside the mind. Given this, the public word 'Harvard', the story would go, is "unreal" in roughly the same way that the sky, the daily sunset, perceived colour, and rainbows are not real considered as external objects. In all these cases, we project "out there" something that is really an amalgam of things going on inside the mind, and (non-obvious) things that are going on in the external world: "the structure of language is not "out in the world but [is] rather a consequence of the mental organization of language users" (Jackendoff, 1987: 133). Ordinary people cannot fail to think of the sky, the sunset, blueness, and rainbows as mind-external objects, wholly out in the world, even after careful scientific training. But what scientific investigation teaches is that, appearances notwithstanding, they are partly in the individual mind. (Importantly, being open to taking these results seriously, thereby setting aside common sense, is part and parcel of being a methodological naturalist.) The illusion that there really are public words, words "out there" that we share, is reinforced by the fact that people talk about words. An egregious case in point, as noted: the very argument against the existence of words apparently used as premises claims about words. But, coming to the second reply, that we talk about, say, 'Harvard', does not actually entail that there is a public word "out there" that we share. On the one hand, speakers regularly refer to things that simply do not exist: Santa, unicorns, the present King of France, etc.1 On the other hand, even if there are some unquestionably real things that we refer to, when we speak of the word 'Harvard', there needn't be a single object which is the public word. A plausible alternative view is that there are many, many words 'Haward'. For some purposes, we count all pronunciations as constituting "the word 'Harvard' "; for other purposes, we count only very few. And so on. We refer to different sets on different occasions, depending upon the context. The resulting sets are real, and they are intersubjective. Sta, there isn't one thing, the word 'Haward'. (See Bilgrami 2002 for this general line of thought.) Hence we can consistently talk about "the different pronunciations of 'Harvard' ", without committing ourselves to there being one unique thing, that publicly shared word, that can stand in a refers-to relation. If the foregoing considerations work, then there cannot be a comprehensive sci- Indeed, what makes this the "radical variant" is that if this criticism is success 'O Put rnetalinguistically, reports of speaker reference are referentially opaque in a way that ex^ reference, if it existed, would not be. See Bencivenga, 1983 for extended discussion; see also Jack 1987: 127 and Chomsky, 1995b: 150.
11 MEANING AND REFERENCE: SOME CHOMSKIAN THEMES not only can there be no comprehensive science of word-world relations, there can be no truths of any kind that state relations between public words and worldly entities. (How could there be, if there aren't any public words?) The Moderate "Argument from Ontology" (Science and Common Sense) The first "argument from ontology" involved arguing that public words don't exist at all. The second "argument from ontology" accepts the reality of both relata. But it questions whether there can be a genuine science of the kind of common sense objects involved on both sides of the relation." Crucial to the argument wiu be the contrast between the world revealed to us by common sense, and that revealed by modern scientific inquiry. We therefore need a way to draw that distinction. The methodological naturalist thinks there is a way to draw it, given nativism and modularity: we can distinguish the world revealed by common sense from the world revealed by science cognitively, in terms of the kinds of concepts deployed. The concepts of common sense, in the sense intended here: are not social artifacts, but are rather part of our biological endowment; more precisely, they are constructed from innately given semantic features-though only the elements out of which the concepts are constructed are innate (and universal), not the resulting wholes; (See Chomsky 2000b: 185.) are acquired (rather than learned), and they do not need to be taught-indeed, given the poverty of stimulus, it's unlikely that they could be learned byitaught to a creature lacking the requisite innate endowment; are at the disposal of every non-pathological human; bring with them a rich and complex internal structure that eschews elegance in favor of day-to-day practicality, especially for living in human company-precisely because they are built out of an innately given store of features; have, finally, and related to this last point, inherently built in implicit references to human hierarchies, rightsiobligations, and our intentional states, rather than aiming for an objective description of the world, independent of us. Scientific concepts, in sharp contrast, are social artifacts.12 More than that, a useful scientific concept is often a hard-won achievement of many years of collective labor. Such concepts must be taught; and frequently enough they cannot be learned, even by non-pathological people. Their content is austere, rather than rich. And, far from l1 That the issue is a science of public words, and public languages, is missed by some of Chomsky's critics. See, for example, Wiggins, l2 I am unsure whether Chomsky himself would endorse what follows. He sometimes suggests that humans have a "science forming faculty", and if scientific concepts derive from it then they are not especially social after all. Since Chomsky exegesis is not my aim, however, I leave this issue aside here. (Thanks are due to Julius Moravcsik for drawing my anention to the issue.)
12 924 ROBERT J. STAINTON being tied to anthropocentric interests, the whole idea of a scientific concept is to capture how things "really are" independent of us. It is telling, too, that science involves explicit reflection not just on the concepts it creates to describe and explain, but also on what counts as good evidence, justification, etc. Those standards of evaluation too are sanctioned by groups, over extended periods of time; they aren't just "given by nature". Sciences, and scientific concepts, are thus artifacts of a social practice, rather than being innately specified-which allows, as Chomsky (1993: 32) suggests, that science can afford to disregard common sense, and is happy to move beyond some ofits tenets. The overall picture can be summed up with the following long passage from Chomsky's Language and Thought: We have, by now, fairly substantial evidence that one of the components of the mind-brain is a language faculty, dedicated to language and its use-where by "language", now, we mean human language, not various metaphoric extensions of the term. Other components provide "common sense understanding" of the world and our place in it... Other components make it possible for humans to conduct scientific and mathematical inquiry, and sometimes to achieve remarkable insight: we may call them "the science-forming faculty", to dignify ignorance with a title. These could be quite different in character from those that yield "common sense understanding" in its various forms. It is an open empirical question, and no dogmatism is in order. The history of modem science perhaps suggests that the distinctions are not trivial; at least, that is one way to interpret the startling conflicts that have arisen between common sense understanding and what scientific inquiry reveals. (Chomsky 1993: 34-5) Having contrasted science and common sense, we can now note that the concepts fall into two families, and add that each collectively provides a perspective.13 What does the "scientific perspective" show us? Quarks, tectonic plates, genomes, and many other things. (To be clear, the scientific perspective does not merely reveal socalled "physical" objects; the mind-brain, at various levels of abstraction, can also be seen from this perspective. That, indeed, is just what linguistics is supposed to help reveal. And, of course, the common sense perspective does not merely reveal psychological states: it affords views of desks, house pets, tea, and toys.) Importantly, however, there is lots that the scientific perspective does not show us: it does not encompass normative categories like good wine, liveable cities, or well-prepared osso bucco; nor does it even encompass not-explicitly normative yet mind-dependent things like clouds,i4 tea,15 desks, sunsets, breakfast cereal, and hockey scores. In I' Chomsky (1993: 48) writes: "The information provided by lexical items and other expressions yields for thinking and speaking about the world..." Or again, "a lexical item provides us with a certain range of perspectives for viewing what we take to be things in the world, or what we conceive in other ways; these items are like filters or lenses providing ways of looking at things and thinking about the products of our minds" (Chomsky, 1992a: 36). l4 On the natural assumption that whether a quantity of water in the atmosphere is a cloud depends upon mind-dependent relations-lie being visible, in normal circumstances, to the naked eye of normal humans. 'j As Chomsky has frequently noted, what is chemically the very same substance could be tea-if created by dipping a bag of tea leaves into a cup of hot water-or contaminated water-if created, say,
13 MEANING AND REFERENCE: SOME CHOMSKIAN THEMES gneral, being objective and ignoring interest-relative distinctions, the "scientific cannot see entities whose individuation conditions inherently involve complex human interests and purposes. Those things are only "seen" through the inborn lens of natural language expressions and the sorts of sub-lexical concepts that help make up these expressions.16 The reason why the scientific perspective cannot "see" such objects is most easily illustrated with examples. Take London. Chomsky (1993: 22-3) writes: We can regard London with or without regard to its population: from one point of view, it is the same city if its people desert it; from another, we can say that London came to have a harsher feel to it through the Thatcher years, a comment on how people act and live. Referring to London, we can be talking about a location, people who sometimes live there, the air above (but not too high), buildings, institutions, etc., in various combinations. A single occurrence of the term can serve all these functions simultaneously, as when I say that London is so unhappy, ugly, and polluted that it should be destroyed and rebuilt 100 miles away. As Chomsky goes on to conclude, "No object in the world could have this collection of properties" (1993: 23). To be clear, it is not being claimed that London simply fails to exist. "London is not a fiction" (Chomsky, 1992a: 37). This is a key difference between the radical worry in the last section, about public words, and the present moderate worry. The issue this time is not whether the relata exist, but (as it were) what makes them exist: the worry is that the kind of socially constructed object that is London (and 'London'!), so highly dependent on human perspectives and interests as it is, cannot be seen by the peculiar instrument that is natural science. This, and not the very existence of London, is what seems doubtful. To generalize the point, imagine natural science looking for external world correlates of 'bargain', 'ownership', 'tenure', 'delicious', 'sacred', 'funny', 'notary public', 'nearby' or 'polite'. How can genuine science see any of these? The thing is, if many, or most, or even all common sense concepts are covertly like 'London' and the rest-and, given the contrast between scientific concepts and common sense concepts that emerged above, this is just what seems to be the case-then a scientific reference-based semantics is hopeless for the vast majority of lexical items in natural language. Crucially for present purposes, granting that what common sense "sees" is perfectly real, we still arrive at the conclusion introduced at the outset: that a comprehensive science of language cannot (and should not try to) describe relations of semantic reference, i.e. word-world relations. That is because the things which manage to be, on this more moderate view-i.e. both words and ordinary objects-are by the right combination of organic matter falling into a lake. Because "what is tea" reflects human interests in this way, science isn't in the business of contrasting tea from non-tea. See Chomsky, 1995b: 128 and Chomsky, 2000b: 189 for discussion. l6 Some of the authors discussed here-e.g. Jackendoff and McGilvray-take an additional neo- Kantian step, and draw the metaphysical conclusion that there are two "worlds", one for each perspective. But this is not an immediate consequence of the present argument about the nature of scientific semantics. One can resist the metaphysical step, and stick to the epistemological claim that the scientific perspective can only "see" part of the one world. Either approach will yield the desired conclusion.
14 - 926 ROBERT J. STAINTON nevertheless not real in the right sort of way. Hence they cannot be "seen" from the scientific perspective. They are as "real" as governments, townships, by-laws, nearness, corporations, national dishes, and so on-which, as Heidegger and G. E. Moore rightly insisted, are as real as can be-but, like townships and nearness, they are invisible to the working scientist. In particular, then, they are invisible to the scientist of language, as such. Before continuing, I should consider some natural objections. As a reminder, I have been trying to contrast two ways in which to look at the world: from the perspective of common sense, and from the perspective of modern science. I did so by contrasting two families of concepts, namely common sense concepts, and scientific concepts. I then suggested that, from the scientific perspective, there are many things which just cannot be seen-and hence that no science of language can describe them. Consider now four objections to this line of argument. First, one might object that the distinctions between the kinds of concepts are being drawn, at least in part, along lines that a radical empiricist or cognitive holist would question. The reply to this is obvious. As hinted right at the outset, methodological naturalists consider radical empiricism, and also anti-modularity, to be empirical theses. They also consider modularity and a very significant innate endowment for language to be well supported on empirical grounds. So, the fact that the case against reference-based semantics may rest on such hypotheses will not detain them. Second, one might reasonably complain that there will be a host of potential referents which the natural sciences can see, and which therefore could be described in a theory of word-world relations (assuming public words to exist). These are precisely the objects that the perspective of science specifically affords: quarks, tectonic plates, wh-traces, etc. Thus the foregoing argument does not on its own rule out a "linguistic science of reference" for the vocabulary of the sciences. (See Chomsky, 1992a: 42-3 and Chomsky, 1993: 27 ff for discussion.) Granted, one might not be able to have a comprehensive science of language; but a limited reference-based scientific semantics would still be possible. However, this is ruled out independently, given the Chomskian view that the proper object of study, for a science of language, is the human language faculty. That, after all, is the aspect of reality that we find, when we start to "investigate language" naturalistically. Now, the language faculty is the innately given part of the mind-brain which, in response to environmental triggers, settles into a steady state of linguistic competence sometime before age five; crucially, then, not everything one "learns about language", as we pre-theoretically say, belongs in the language faculty. In particular, Chomsky suggests that scientific vocabulary likely is not stored in the language faculty: it is learned, not acquired; it is austere, not rich; it is a construct, not an aspect of our biology. Thus, while there might conceivably be a science of reference for scientific terms, it would not be part of the science of natural '' An interesting side effect of distinguishing common sense concepts from scientific ones is that one could follow Quine in holding that there is no andyticlsynthetic distinction for the concepts used in
15 Third, one might doubt that science and common sense really do provide mutually exclusive and exhaustive perspectives. First, it's clearly true that the supposed gulf between science and common sense seems bridgeable in places, and this on at least three grounds: (a) there seem to be factual claims that straddle the divide: e.g. 'My cup of herbal tea boiled at degrees centigrade'; (b) common sense reports can clearly serve as evidence forlagainst a scientific hypothesis: e.g. 'The Prime Minister died after eating one of those' or 'It turned red like a fire truck when we poured the liquid on it' can both support (or call into question) a genuine scientific hypothesis about the chemical make up of an unknown item; (c) it's arguable that there are whole disciplines which straddle the borderline: criminology, epidemiology, anthropology, medicine, horticulture, etc. Thus the difference may be not a difference of kind, but one of degree. (See Moravcsik, 1998: 127 for extended discussion.) This is a very important objection. It therefore merits a detailed rebuttal, comprising three related replies. The first reply is that there cannot be a comprehensive science of language unless that discipline can see all (or at least the vast majority) of objects that can bear names. Hence the conclusion stands even if the difference between what common sense can see, and what modem science can see, is a matter of degree, with intermediate cases along the way-as long as there are lots of things on the extreme end that science can't see. In short, the conclusion argued for doesn't really require the stronger premise that the distinction is exclusive and exhaustive. The second reply makes a related point about making do with a weaker premise. What the conclusion requires is not really the claim that "science simply cannot see common sense objects"; what it requires, instead, is merely that there is no single science which can see (almost) every common sense object-since a comprehensive science of language that described word-world relations would need to be just such a science. To imagine a comprehensive science of language that posits word-world relations is, for instance, to envisage a single genuine explanatory science that can "see" all of friends, yarmulkes, Tuesdays, symphonies, jokes, vagrants, bargains, and every other common sense objea.l8 Put another way, the second reply is this: what is required, at a minimum, for a comprehensive scientific semantics that introduces word-world relations is There exists a science x such that, for almost every y, x can see y; but at best what is plausible is For almost every y, there exists a science.x, such that x can see y. The third reply to this third objection is that even this weaker claim (which wouldn't actually avoid the conclusion in any case) gets much of its plausibility from loose terminology. To repeat, as the term science (slnce those words really do get their meaning holistically, from their place in a world-describing theory), while nevertheless insisting that common sense concepts-built as they are from innately specified features-will license analytic truths. See Chomsky, 2000b: 186. For a discussion some of the epistemological implications of this way of slicing things, see Matheson and Stainton, (2002). l8 Jackendoff draws a weaker (but stdl very interesting) conclusion, on related grounds. He argues that "language is about ent~ties in the world as construed by the language user/perceivern (1987: 128). (See also Jackendoff, 1991: 12.) As a result, word-world relations cannot be studied prior to, and independent of, psychological investigations about how humans categorize. In which case, external-referent semantics cannot be a genuine alternative to Jackendoff-style Conceptual Semantics, since the former implicitly presupposes the latter.
16 928 ROBERT J. STAINTON - is being used here, not every systematic empirical inquiry counts as 'science': e.g. an exhaustive taxonomy of the Earth's beetles is not science, in the sense I have been discussing-not least because mere taxonomy does not seek out underlying explanatory realities. Noting this, it's not even clear that most common sense objects are "seen" by any genuine science after all, though it might well be plausible that some systematic empirical inquiry can see each such object. Whereas the three previous objections were specific to the moderate argument from ontology-each attacking in a different way the proposed sharp and exhaustive contrast between the perspective of common sense and the perspective of modern science-the final objection, which is methodological in character, applies more globally to both arguments from ontology. The complaint here is that even making these two ontology based objections flies in the face, paradoxically enough, of what I described as the core tenet of the Chomskian approach to language, viz. methodological naturalism. From that perspective, the test of a theory is, surely, whether it can establish a rich body of empirically supported doctrine; and, the fourth objection continues, semantics has made great progress in those terms. There are thick textbooks full of results, journals packed with data and detailed debates, and so on. Thus, the fact that reference-based semantics might fail to meet certain arch ontological scruples should not cause us to reject it, since it is a thriving research program. Actually, Chomsky himself agrees with the methodological point: questions of ontology are, for him, posterior to questions about explanatory and descriptive success (Chomsky, 2000b: 184). Presumably every methodological naturalist will agree. The objection still misfires, however, because it is based on the (thoroughly mistaken) idea that "Chomskians" reject semantics root and branch. It's not the entire sub-discipline of semantics that is being rejected, only a peculiar spin on it. And, continues the reply, the many existing results of semantics have little or nothing to do with the extraneous philosophical hypotheses that shared public words/sentences stand for real-world external objects, sets thereof, and world-bound thoughts.lg Indeed, meaning broadly construed remains as central to Chomskian linguistic theorizing in 2003 as it was in To give but one example, Chomsky's most recent Minimalist theorizing makes essential use of the principle of Full Interpretation, which requires (among other things) that only elements that have an interpretation can remain at the end of a derivation-this being the point where l9 Some would say that, beyond leaving the key results of formal semantics standing, absolutely nothing is lost by linguistics, when it abandons the reference-based approach; moreover, much is gained. Thus Hornstein (1989) writes: "If semantics is concerned with truth conditions, and this is construed as correspondence, then I can see no reason for thinking that there is any link between semantics so defined and theories of linguistic interpretive competence. Moreover, this is all for the good as far as the latter enterprise is concerned, for semantic theories seem to require the ascription of powers and capacities to native speakers which are as mysterious as those capacities that we wish to explain. Syntactic theories, those types of theories that eschew language-world relations, are not similarly problematic. It is for this reason that syntactic theories are methodologically preferable". See also scattered remarks in Hornstein 20 Neil Smith ( 1999: 163) quotes Chomsky as saying that "putting aside phonology, virtually everything I've done in the field since LSLT [The Logical Structure of Linguistic Theory] falls within semantics". See also Chomsky, 2000b: 174.
17 - MEANING AND REFERENCE: SOME CHOMSKIAN THEMES the string reaches the interface with Conceptual Structure. AU other structure (e.g. and accusative case marking, expletives, agreement features not bearing ont tent) must be "checked", thereby being rendered "invisible", before this point. Thus meaning couldn't be more central to current linguistic theory. (See Radford, 1997: 170ff for introductory discussion, andchomsky, 2001; Chomsky, 2000~; Chomsky, 1995a for the evolving details of Minimalism.) Moreover, as Pietroski (2003, 2005) points out, there is much of the same positive hard work for the subdiscipline of semantics to do, without reference and truth. First, this sub-discipline will explore semantic properties of expressions such as: what they can and cannot mean; whether they are ambiguous; if they are ambiguous, why, if they are not, why not; what referential dependencies must, can and cannot obtain; and so on. In fact, for all that has been said here, semantic theory could even retain the architecture of a Fregean or Tarskian theory, with both primitives and complexes, lexical and compositional rules, different semantic types, functions combining with arguments, compositional determination of whole-meanings from part meanings, and so on.*l What is rejected is just the idea that the primitives stand for real world objects and sets outside the mind, and that sentences express world-bound thoughts (relative to a set of parameters). Thus, the two objections from ontology in no way force one to abandon the discipline of semantics, or its many results. (We will revisit the tasks of semantics at the end of the chapter, when discussing the positive alternative to reference-based semantics.) The Failure of Compositional Referential Semantics As presented here, the former two arguments have focused primarily on public words (e.g. 'Hanard', 'photographer') and external objects (e.g. London). But the Chomskian target is the whole tradition of treating natural languages as, in key respects, like the formal languages invented by logicians-and, as explained above, that tradition brings with it aview not just about words and objects, but also views about predicates (e.g. verbs and adjectives) and sets, and sentences and truth. Traditionally, sentences are assigned "proto-thoughts": things which, given a particular setting for a fixed cluster of parameters (time, place, speaker, addressee), are true or false. And predicates are assigned sets of objects as their extension. These ideas are equally questionable, says the Chomskian-even putting aside the issues, raised above, of whether there are "public sentences" and "public predicates", and whether sets of common sense objects and proto-thoughts can be "seen" from the scientific perspective. As noted at the outset, the mark of a Chomskian, as I here intend the term, is a commitment to methodological naturalism. Now, the methodological naturalist, 21 See Jackendoff, 1983, 1991, 1996, 2002 for detailed examples of keeping much of this structure, but without external reference based semantics. It is worth stressing, however, that some of the authors discussed here remain highly skeptical about retaining this traditional superstructure, within a naturalist framework. See in particular Moravcsik 1998, who maintains that Fregean and Tarskian systems (i) need sharply defined word meanings, not permitting polysemy, and (ii) require that syntax mirror semantics-neither of which conditions, Moravcsik argues, hold for natural languages.
18 inquiring into language as an aspect of nature, will follow the canons of the sciences, and will seek out, as her object of inquiry, a real object-possibly differing in important ways from pre-theoretical conceptions-that is scientifically tractable. That object turned out to be the language faculq. that innately provided, specialized module of the mind-brain. For this reason, the methodological naturalist will approach semantics, like phonology and syntax, as part of the study of the language faculty. What will emerge below, however, is that it is not plausible that the language faculty, taken alone, can assign proto-thoughts to sentences, or sets of objects to verbs, adjectives, etc. Hence sentence meanings cannot be thoughts, nor even "thoughts relative to a set of parameters"; and predicate meanings cannot be sets of objects in the external Let's begin with sentences. The fundamental points here are made by Pietroski (2003, 2005), building onchomsky (1977). On the one hand, there is no empirical reason for thinking that what the language faculty assigns to a sentence wdl be capable of being true or false, even given contextual parameters like time, place, speaker, hearer, etc. (There's lots of empirical reason for thinking that people can say, and think, things that are true or false; but that is another matter.) The only thing which drives one to this expectation is, at bottom, a dubious analogy between natural objects and artifacts whose properties are stipulated (e.g. the predicate calculus). For the methodological naturalist, that in itself is damning. On the other hand, there is lots of empirical evidence that the language faculty alone doesn't assign thoughts (or propositions, or truth conditions, or what have you). In particular, very many sentences either lack truth conditions altogether, or are assigned truth conditions only via the rich interaction of different mental faculties. Consider, for instance, the following pair of sentences: (1) Poems are written by fools like me (2) Mountains are climbed by fools like me Putting aside the difficult question of what a referential semantics would assign as meanings for the parts (e.g. what real-world object does the plural word 'poems' stand for?), it does seem that the same kind of meaning, whatever it is, would have to be assigned by the language faculty to 'poems' and 'mountains', and to 'are written' and 'are climbed'. Moreover, the same syntactic structure appears in both sentences. Thus, the prediction would be that, as far as the language faculty goes, (1) says about poems and being-written whatever (2) says about mountains and being-climbed. Yet, insofar as they assign a proto-thought at all, the proto-thought that an agent would typically associate with (1) requires that all poems are written by fools, while 22 Looked at slightly differently, as Fodor (2001) has argued, if semantics has to yield "a thought expressed (give or take a bit)", then a compositional semantics for natural language is just not ~ossible. What one should conclude, says the Chornskian, is not that natural languages lack compositional semantics, which is Fodor's conclusion, but rather that scientific semantics just shouldn't be in the business of assigning thoughts to sentences-nor even "thoughts give or take a bit". Instead, the linguistic meanings of whole sentences just are those thiigs-whatever they turn out to be-which are compositionally determined from part-meanings plus syntax.
19 he proto-th~~ght that an agent would typically associate with (2) does not require, for its truth, that all mountains are climbed by fools. Thus the truth conditions hat language users tend to assign are importantly different. Nor is this an isolated example. Think, for instance, of how knowledge of the world impacts on what sense one assigns to 'may' in 'Marta may get cancer' versus 'Marta may smile if she wishes to.' In so far as one treats these as truth evaluable at all, one hears the first 'may' as expressing (epistemic or physical) possibility, and the latter 'may' as expressing pemissibility. Moreover, this contrast in how 'may' is understood arises because, as a matter of fact, one isn't given permission to get cancer; and because it's too obvious to bear mention that it's (physically and epistemically) possible for a person to smile, if she wishes to. Or again, what the concatenation of a nominal and modifier contributes to meaning varies widely from case to case; in particular, the meaning of the nominal-modifier complex frequently reflects facts known about the world. For instance, compare 'Christmas cookie' ("made to be consumed at"), 'Girl guide cookie' ("sold by"), 'oatmeal cookie' ("made of"), 'yellow cookie' ("coloured"), 'fortune cookie' ("containing"), 'doggie cookie' ("made to be eaten by"), and 'Walrnart cookie' ("sold at"). Given that concatenation doesn't always mean the same thing, if an agent is able to assign truth conditions at all to 'Phyllis ate a Boy Scout cookie', it is because the agent knows the relevant facts about Boy Scouts: e.g. that they are not an appropriate ingredient for a cookie, that there isn't a Boy Scout holiday, that Boy Scouts are too big to be inside a cookie, etc. One might reply that language users are simply poor judges of what sentences mean: "the folk" mix up what the expression means with how it would be standardly used. For example, while language users do indeed assign quantificationally different truth conditions to the sentence types (1) and (2), they are wrong to do so. This is a fair point. Indeed, Chomskians themselves are wont to note that language users do not have direct insight into the syntax of their own idiolect: language users are apt on first hearing, for instance, to incorrectly judge that 'The horse raced past the barn fell' is ungrammatical. Still, if we ask the semantic theorist what the proto-thought expressed by (1) is, since by hypothesis it isn't that all poems are written by fools like me, it will become clear that the meaning that the language faculty taken alone assigns is not the sort of thing which, even relative to a set of parameters (speaker, addressee, time, place), is true or false. Similarly for 'Marta may get cancer' and 'Phyllis ate a Boy Scout cookie.' Whatever the context-invariant meaning of these sentences is, it's something much more abstract than a thought, or even a function from a restricted set of parameters to a tho~ght.~~ Indeed, this attempted reply brings out even more clearly the deeper problem with trying to assign truth conditions to sentences. That deeper problem has to do with where the differences in truth conditions that agents assign-e.g. between (1) and (2)-derive from. Part l3 The point is, of course, closely related to the idea-defended by Robyn Carston, Francois Rkcanati, John Searle, Dan Sperber and Deirdre Wilson, and Charles Travis-that there are "pragmatic determinants" of the truth conditions of speech acts. See Carston 2002 for detailed and Uuminating
20 932 ROBERT J. STAINTON - of the Chomskian point is precisely that the difference in the truth conditions that language users assign, in so far as they do assign truth conditions, derives from real world knowledge that people have-e.g. about poems versus mountains (i.e. that the former are all human creations, but the latter are not). Moreover, there is no other source of truth conditions: leave that real-world knowledge out and what is detemined is too abstract to bear a truth value. Thus, in so far as we agents assign truth conditions to sentences at all, the truth conditions we assign are a massive interaction effect of different kinds of knowledge: knowledge afforded by the language faculty, yes, but also knowledge afforded by many other parts of the mind-brain. Not being solely an aspect of the language faculty, it follows that the truth conditions which people assign to sentences do not fall within the domain of the science of language. (See Borg, this volume, for a rather different view.) Another kind of case raises problems for the idea that predicates (verbs, verb phrases, adjectives, adjective phrases, etc.) have sets of external objects as their content. Compare sentences (3) through (6): (3) The house is green (4) The ink is green (5) The banana is green (6) The stoplight is green In each case, in so far as talk of "contributing sets" is appropriate at all, is green] appears to be contributing a quite different set in the four cases. In (3), the house must be in the set of things which are green on the outside (though the house need not be entirely green on the outside). Similarly for (5), which requires only that the banana peel be green. In contrast (4) requires that the stuff (which right now looks black) be in the set of things which, when applied to paper and allowed to dry, will be green. As for (6), the science of colour tells us that the property exhibited by the stoplight is physically very different from that exhibited by the banana peel in (5): the stoplight being green involves not the reflection of light, but the emission of light. So, thought of as a physical set, the one which [Iis green] picks out in (6) is very different yet again. (See: Moravcsik 1998: 44-5 for similar remarks about 'is white'; Jackendoff, 1991: 44 on different senses for 'end'; and Jackendoff, 1983, 2002 and Pustejovsky, 1995 for a panoply of other examples.) Part of what appears to be going on here is this: which set of things is associated by speakers with one part of the sentence depends upon what they associate with the other parts. Here, the set that speakers associate with [P is green] depends on the kind of thing that 'green' is thought of as applied to: houses, ink, bananas, etc. Worse, the variation in the set selected by the agent as the denotation for 'green', depends upon facts about how reasonable speakers would use the sentence-which in turn depends on factors like how likely it is for houses to be wholly and completely green, what ink is used for, etc. Once again, then, the denotation, in so far as language users assign one at all, is an enormous interaction effect, and does not depend solely upon the language faculty; so, there can be no science of language which assigns sets to predicates.